How can I make small super computer with VMW DRS? NO GPU!!! Forget this scenario.
I have 10 servers with DUAL EPYC CPU 2x64 cores. This mean I have 1,280 cores.
What are the steps to do that?
1. Install ESXi on each of 10 servers.
2. Install VM Data Center or vCenter???
2.1. add all ESXi servers into a laboratory
2.2. then add all ESXi server into a single cluster
2.3. activate DRS
3.. Install Windows IIS on each of ESXi (Virtualization Windows OS)
3.1. Deploy Microservices on each of IIS
3.2 Install RabitMQ on a machine (into windows)
3.3. Deploy Microservice Gateway into Windows (let assume in physically machine on first ESXi )
4. Test
4,1 Send 1M ( one millions) request to Gateway.
Am I missing something?
Do you want to see the result? Analytics? What really happing? How VMW DRS will move all cores to first ESXi?
Now let make a new scenario. Delete VMW Fabric and do this with RabitMQ only? Do you want to see the result? Analytics? What really happing?
Virtual CPUs per virtual machine (Virtual SMP)
768
Moderator: Moved to vSphere Discussions - the {code} area where you created this thread is for SDK matters.
Well what you are missing I guess is the fact that DRS doesn't enable you to create one large super computer. It provides the ability to create a "cluster of resources" which provides loadbalancing. It doesn't aggregate all the resources for you to assign to a single Virtual Machine which are then scheduled across physical hosts. It doesn't work like that unfortunately.
Hm... The old version of DRS using average algorithm. The current version of DRS move resources from unused machines to machine witch is working intensive operation. This seems like small super computer.
Can I forcedly set some options and get all resources from 99% of machines into one?
VMware Bitfusion with GPUs??? https://youtu.be/VFuIpheXpyA
Apart of that gpu options have also some config bug https://youtu.be/w-Ez4Bqt33s
@depping wrote:Well what you are missing I guess is the fact that DRS doesn't enable you to create one large super computer. It provides the ability to create a "cluster of resources" which provides loadbalancing. It doesn't aggregate all the resources for you to assign to a single Virtual Machine which are then scheduled across physical hosts. It doesn't work like that unfortunately.
Besides the obvious reasons Duncan mentioned, why do you want to put all your physical resources into a single box as your design paradigm is focused on delivering microservices? That is a distributed paradigm